Knight

The job application layer you've been missing.

Every application is the same form. Different URL, same twenty fields — name, email, phone, work history, education — pasted by hand into Workday, Greenhouse, Lever, and a dozen other portals that share nothing between them. Most autofill tools break on anything beyond a basic text input, or quietly ship your resume to a hosted AI pipeline to do the parsing.

Knight fixes this without the tradeoffs. Your resume is parsed entirely on your own machine by a local Python service. Field mapping runs through whichever LLM you choose — local Ollama or a cloud provider you configure. Before a single character is written to any form, you see a diff of every proposed change and confirm it. Your data stays on your device.

🌐 sir-ad.github.io/knight

---

Features

Capability	Detail
Resume parsing	PDF (digital + scanned OCR), DOCX, TXT, or pasted text
Local-first	PyMuPDF + optional Tesseract OCR in a local Python sidecar
Autofill	14 named ATS adapters + generic DOM fallback
Review overlay	Inline diff of what will be filled — confirm before anything is written
Application log	Local status history with auto-detection from Gmail
Gmail sync	Read-only OAuth — spots confirmations, interviews, rejections, offers
Follow-up drafts	Generates follow-up emails from your application context
LLM providers	Ollama · OpenAI · Anthropic · Gemini · OpenRouter

---

Architecture

┌──────────────────────────────────────────┐
│           Chrome Extension (MV3)         │
│                                          │
│  ┌──────────┐  ┌─────────┐  ┌────────┐  │
│  │ Service  │  │ Content │  │ Popup  │  │
│  │ Worker   │◄─│ Script  │  │  UI   │  │
│  │ (bus +   │  │(inject/ │  │(React) │  │
│  │ LLM hub) │  │ fill)   │  │        │  │
│  └────┬─────┘  └─────────┘  └────────┘  │
└───────┼──────────────────────────────────┘
        │ HTTP (localhost only)
        ▼
┌───────────────────────┐
│  Resume Parser Sidecar│  http://127.0.0.1:43118
│  FastAPI · Python     │
│  PyMuPDF · Tesseract  │
│  LangExtract          │
└───────────────────────┘

All LLM calls from the extension route through the service worker — content scripts never call APIs directly. The sidecar handles heavy file parsing outside the browser sandbox.

---

Supported ATS Portals

Portal	Match pattern
Workday	*.myworkdayjobs.com, *.workday.com
Greenhouse	boards.greenhouse.io, *.greenhouse.io
Lever	jobs.lever.co, *.lever.co
Naukri	*.naukri.com
iCIMS	*.icims.com
SmartRecruiters	*.smartrecruiters.com
Oracle Taleo	*.taleo.net
SAP SuccessFactors	*.successfactors.com, *.successfactors.eu
LinkedIn Easy Apply	*.linkedin.com/jobs/*
Indeed	*.indeed.com/apply/*
Wellfound / AngelList	*.wellfound.com/jobs/*, *.angel.co/jobs/*
Ashby	*.ashbyhq.com
Rippling	*.rippling.com/jobs/*
Generic fallback	Any form with standard labels, placeholders, or aria-label metadata

---

Quick Start

One-line install

bash <(curl -fsSL https://raw.githubusercontent.com/sir-ad/knight/main/install.sh)

Clones the repo, builds the extension, and sets up the Python sidecar in one shot. Then follow the printed instructions to load the unpacked extension in Chrome.

Set KNIGHT_DIR to change the install location (default: ~/knight).

---

Manual setup

1. Start the resume parser sidecar

git clone https://github.com/sir-ad/knight.git
cd knight/resume-parser-sidecar
./scripts/setup-venv.sh   # one-time: creates .venv and installs deps
./scripts/run.sh          # starts on http://127.0.0.1:43118

For scanned PDF support, install Tesseract:

# macOS
brew install tesseract

# Ubuntu / Debian
sudo apt-get install -y tesseract-ocr

2. Build the extension

cd ../careerflow
npm install
npm run build

3. Load in Chrome

1. Open chrome://extensions
2. Enable Developer Mode
3. Click Load unpacked
4. Select careerflow/build/chrome-mv3-prod

4. (Optional) Start Ollama for local LLM

OLLAMA_ORIGINS=chrome-extension://* ollama serve
ollama pull llama3.2:3b

---

Requirements

Component	Requirement
Node.js	20+
Python	3.11+
Chrome	115+ (MV3)
Ollama	Any recent version — for the local-first path
Tesseract	Optional — only needed for scanned PDF resumes

---

LLM Provider Setup

Knight uses the provider you configure in Settings. Ollama is the default and requires no API key.

Provider	Key required	Notes
Ollama	No	Default. Runs 100% locally.
OpenAI	Yes	Supports GPT-4o, GPT-4.1 series
Anthropic	Yes	Supports claude-haiku-3-5, claude-sonnet-4-5
Google Gemini	Yes	Supports gemini-2.0-flash
OpenRouter	Yes	Routes to any supported model

Store keys in Settings → the extension keeps them in local Chrome storage and never includes them in exports.

Ollama endpoint: enter the host root only — http://localhost:11434. Do not append /api.

---

Gmail Sync

Knight requests a read-only Gmail OAuth scope (gmail.readonly). It scans for emails from known ATS domains and classifies them as confirmation, interview invite, rejection, or offer — updating your local application log automatically.

To enable Gmail sync, set your OAuth client ID before building:

# careerflow/.env.local
PLASMO_PUBLIC_GOOGLE_CLIENT_ID=your-google-oauth-client-id

---

Development

Extension

cd careerflow
npm run dev          # Plasmo dev server with hot reload
npm run typecheck    # TypeScript check
npm test             # Jest tests
npm test -- --testPathPattern=<file>  # Single file
npm test -- --coverage

Sidecar

cd resume-parser-sidecar
source .venv/bin/activate
python -m uvicorn resume_parser_sidecar.app:app --host 127.0.0.1 --port 43118 --reload

pytest tests -q
pytest tests/test_service.py  # Single file

Release gate (full validation)

# Extension
cd careerflow
npx tsc --noEmit && npm test -- --runInBand && npm run build && npm run package

# Sidecar
cd ../resume-parser-sidecar
source .venv/bin/activate && pytest tests -q

---

Repo Layout

Knight/
├── careerflow/                  # Chrome extension (TypeScript · React · Plasmo)
│   ├── src/background/          # Service worker — message bus, LLM calls, Gmail sync
│   ├── src/content/             # Content scripts — ATS detection, autofill, overlay
│   ├── src/lib/                 # Shared utilities — types, LLM providers, storage
│   └── src/popup/               # React popup — Profile / Dashboard / Settings tabs
├── resume-parser-sidecar/       # Local FastAPI service (Python 3.11+)
│   └── resume_parser_sidecar/   # app, extraction, langextract runner, providers
├── docs/                        # GitHub Pages site
└── .github/workflows/           # CI, release packaging, Pages deploy

---

CI / CD

Workflow	Trigger	What it does
ci.yml	Push / PR to main	typecheck → tests → build → package (extension); pytest (sidecar)
release.yml	Version tags (v*)	Builds and attaches the extension zip to GitHub Releases
pages.yml	Push to main	Deploys docs/ to GitHub Pages

---

License

MIT